An agentic search engine for scientific publications.

Explore the prototype: https://sci-sear.ch/ | made by: Carla Ostmann

Design Question

How might a search engine meet users’ search intents rather than queries?

Overview

In this write-up I document my design process behind making [ sci-search ]

  1. Research I collate learnings from dozens of user interviews conducted over 2 years of working in the researcher tool space into a model of literature search: what it is and why it is. I also summarize my findings from exploring existing tools.
  2. Design Principles Based on an understanding of the use case, existing solutions, and current technological capabilities, I derive a set of design principles to inform the making of [ sci-search ]
  3. Conceptual Model I describe what sci-search is and describe the core user flow schematically.
  4. Prototype UX Following the core user flow through [ sci-search ], I explain how I made the design principles and conceptual model manifest in the “final” prototype.
  5. Closing Remarks Finally, some musings on where to go from here.

Research

The design process was informed by insights gained through dozens of user interviews over the last 2 years as well as first-hand experience with existing scientific search engines. Below I present the key insights that have shapes [ sci-search ]

A model of classic literature search

Fig 1. Keyword-based search yields matching, rather than relevant results. Researchers must codify their search intent as explicit keywords which are taken at face value by the system as relevance criteria. Matching results are presented as a list of paper metadata, leaving it up to the user to determine true relevance and process the information contained within papers for their use.

Fig 1. Keyword-based search yields matching, rather than relevant results. Researchers must codify their search intent as explicit keywords which are taken at face value by the system as relevance criteria. Matching results are presented as a list of paper metadata, leaving it up to the user to determine true relevance and process the information contained within papers for their use.

Search intent != search query

Researchers turn to scientific search engines with the purpose of informing an ongoing research project. Researchers are not searching in order to read someone else’s papers. They’re searching to find the information contained in sections of others’ papers that can advance their own work. In the following paragraphs, I will refer to this latent and idiosyncratic purpose as search intent.

A search intent always exists within the context of an ongoing or planned larger research project.

A search intent always exists within the context of an ongoing or planned larger research project.

A search intent is not a search query. A search intent is “I wonder how psychology and cognitive science are modeling attention - do we know what it is and how it really works?” A search intent is not “attention framework cognition AND psychology.” A search intent lives in the researchers head. And it manifests in online searches just as much as irl and online conversations with colleagues and mentors.*

A search intent is fuzzy and evolves through iteration